equency selective fading channels. However, OFDM systems can be extremely sensitive and vulnerable to synchronization errors. In this paper, we present a scheme for performing timing recovery that includes symbol synchronization and sampling clock synchronization in OFDM systems. The scheme is based on pilot subcarriers. In the scheme, we use a件path time delay estimation method to improve the accuracy of the correlation-based symbol synchronization methods, and use a delay-locked loop (DLL) to do the sampling clock synchronization. It is shown that by using this scheme, the mean square values of the symbol timing estimation error can be decreased by several orders of magnitude compared to the common correlation methods in both the AWGN and multipath fading channels. In addition, the scheme can track the symbol timing drift caused by the sampling clock frequency offsets.

In recent years, combining multiuser detection (MUD) and channel decoding has received considerable attention. The maximum a posteriori (MAP) criterion-based iterative multiuser detector greatly improves the system performance and can approach the performance of single-user coded systems. However, its complexity increases exponentially with the number of users and can become prohibitive for systems with a medium-to-large number of users. In this paper, a reduced complexity MAP-based iterative MUD based on the use of a soft sensitive bits algorithm is proposed for coded multicarrier code-division multiple-access systems. It is shown that it can greatly reduce the computational complexity with a minimal penalty in performance compared to the conventional optimal scheme.

A novcl feature weight (FW) algorithm and an intcgratcd schemc cdled MMSE-FW-LA are proposed. The schcmc combines the minimum mean square error (MMSE)-based enhancemcnt, the FW algorithm and the log-add (LA) modcl compensation. Expcrimcntal cvaluations show that this schcmc can improve thc recognition performance significantly, cspecially at low SNRs.

A new scheme for robust automatic speech recognition is presented. The scheme combines minimum mean square error (MMSE)-based enhancement with Log-Add (LA) model adaptation. Experimental evaluations show that the proposed MMSE-LA scheme can achieve significant improvement in recognition across a wide range of signal-to-noise ratios, especially in very noisy environments.